Ecomon-Copernicus daily resolution species distribution models

This repository encompasses a pipeline for building boosted regression tree (BRT) patch distribution models using NOAA EcoMon abundance data and Copernicus Marine Data Store oceanographic covariates

Output predictions are daily, 1/12° resolution predictions of a species patch from 1993 to 2019 within the Northeast Atlantic.

In addition, the real_time_forecasts folder hosts a Shiny application template for delivering real-time daily product.

Example output: Probability of a dense C. finmarchicus patch daily from April 1st to June 1st, 2015.

Modeling Pipeline & Reproduction

New model versions can be customized in the yaml_maker.R file. Configuration allows for choice of a species; patch threshold; source column within the ecomon::scale_ecomon() output or alternative custom dataset; set of selected copernicus and derived variables; and hyperparameter customization.

Steps:

• Choose a species and input ecomon column. If using a custom abundance dataset, run methods in data_preparation/generate_base_datasets.R accordingly.
• Fill out configuration information and version in yaml_maker.R and execute.
• Update species and version in create_version.R and execute.
• Verify that configuration parameters resulted in adequate model output, either by checking RStudio plot/text outputs or going to model info folder at mnt/ecocast/projectdata/students/ojohnson/copernicus/
• Update species and version in predict_version.R and execute within a shell. Multiple species/version pairs can be run through